4 * Copyright IBM, Corp. 2008
11 * This work is licensed under the terms of the GNU GPL, version 2 or later.
12 * See the COPYING file in the top-level directory.
16 #include <sys/types.h>
17 #include <sys/ioctl.h>
21 #include <linux/kvm.h>
23 #include "qemu-common.h"
24 #include "qemu/atomic.h"
25 #include "qemu/option.h"
26 #include "qemu/config-file.h"
27 #include "sysemu/sysemu.h"
29 #include "hw/pci/msi.h"
30 #include "hw/s390x/adapter.h"
31 #include "exec/gdbstub.h"
32 #include "sysemu/kvm.h"
33 #include "qemu/bswap.h"
34 #include "exec/memory.h"
35 #include "exec/ram_addr.h"
36 #include "exec/address-spaces.h"
37 #include "qemu/event_notifier.h"
40 #include "hw/boards.h"
42 /* This check must be after config-host.h is included */
44 #include <sys/eventfd.h>
47 #ifdef CONFIG_VALGRIND_H
48 #include <valgrind/memcheck.h>
51 /* KVM uses PAGE_SIZE in its definition of COALESCED_MMIO_MAX */
52 #define PAGE_SIZE TARGET_PAGE_SIZE
57 #define DPRINTF(fmt, ...) \
58 do { fprintf(stderr, fmt, ## __VA_ARGS__); } while (0)
60 #define DPRINTF(fmt, ...) \
64 #define KVM_MSI_HASHTAB_SIZE 256
66 typedef struct KVMSlot
69 ram_addr_t memory_size;
75 typedef struct kvm_dirty_log KVMDirtyLog;
84 struct kvm_coalesced_mmio_ring *coalesced_mmio_ring;
85 bool coalesced_flush_in_progress;
86 int broken_set_mem_region;
89 int robust_singlestep;
91 #ifdef KVM_CAP_SET_GUEST_DEBUG
92 struct kvm_sw_breakpoint_head kvm_sw_breakpoints;
98 /* The man page (and posix) say ioctl numbers are signed int, but
99 * they're not. Linux, glibc and *BSD all treat ioctl numbers as
100 * unsigned, and treating them as signed here can break things */
101 unsigned irq_set_ioctl;
102 #ifdef KVM_CAP_IRQ_ROUTING
103 struct kvm_irq_routing *irq_routes;
104 int nr_allocated_irq_routes;
105 uint32_t *used_gsi_bitmap;
106 unsigned int gsi_count;
107 QTAILQ_HEAD(msi_hashtab, KVMMSIRoute) msi_hashtab[KVM_MSI_HASHTAB_SIZE];
113 bool kvm_kernel_irqchip;
114 bool kvm_async_interrupts_allowed;
115 bool kvm_halt_in_kernel_allowed;
116 bool kvm_irqfds_allowed;
117 bool kvm_msi_via_irqfd_allowed;
118 bool kvm_gsi_routing_allowed;
119 bool kvm_gsi_direct_mapping;
121 bool kvm_readonly_mem_allowed;
123 static const KVMCapabilityInfo kvm_required_capabilites[] = {
124 KVM_CAP_INFO(USER_MEMORY),
125 KVM_CAP_INFO(DESTROY_MEMORY_REGION_WORKS),
129 static KVMSlot *kvm_alloc_slot(KVMState *s)
133 for (i = 0; i < s->nr_slots; i++) {
134 if (s->slots[i].memory_size == 0) {
139 fprintf(stderr, "%s: no free slot available\n", __func__);
143 static KVMSlot *kvm_lookup_matching_slot(KVMState *s,
149 for (i = 0; i < s->nr_slots; i++) {
150 KVMSlot *mem = &s->slots[i];
152 if (start_addr == mem->start_addr &&
153 end_addr == mem->start_addr + mem->memory_size) {
162 * Find overlapping slot with lowest start address
164 static KVMSlot *kvm_lookup_overlapping_slot(KVMState *s,
168 KVMSlot *found = NULL;
171 for (i = 0; i < s->nr_slots; i++) {
172 KVMSlot *mem = &s->slots[i];
174 if (mem->memory_size == 0 ||
175 (found && found->start_addr < mem->start_addr)) {
179 if (end_addr > mem->start_addr &&
180 start_addr < mem->start_addr + mem->memory_size) {
188 int kvm_physical_memory_addr_from_host(KVMState *s, void *ram,
193 for (i = 0; i < s->nr_slots; i++) {
194 KVMSlot *mem = &s->slots[i];
196 if (ram >= mem->ram && ram < mem->ram + mem->memory_size) {
197 *phys_addr = mem->start_addr + (ram - mem->ram);
205 static int kvm_set_user_memory_region(KVMState *s, KVMSlot *slot)
207 struct kvm_userspace_memory_region mem;
209 mem.slot = slot->slot;
210 mem.guest_phys_addr = slot->start_addr;
211 mem.userspace_addr = (unsigned long)slot->ram;
212 mem.flags = slot->flags;
213 if (s->migration_log) {
214 mem.flags |= KVM_MEM_LOG_DIRTY_PAGES;
217 if (slot->memory_size && mem.flags & KVM_MEM_READONLY) {
218 /* Set the slot size to 0 before setting the slot to the desired
219 * value. This is needed based on KVM commit 75d61fbc. */
221 kvm_vm_ioctl(s, KVM_SET_USER_MEMORY_REGION, &mem);
223 mem.memory_size = slot->memory_size;
224 return kvm_vm_ioctl(s, KVM_SET_USER_MEMORY_REGION, &mem);
227 int kvm_init_vcpu(CPUState *cpu)
229 KVMState *s = kvm_state;
233 DPRINTF("kvm_init_vcpu\n");
235 ret = kvm_vm_ioctl(s, KVM_CREATE_VCPU, (void *)kvm_arch_vcpu_id(cpu));
237 DPRINTF("kvm_create_vcpu failed\n");
243 cpu->kvm_vcpu_dirty = true;
245 mmap_size = kvm_ioctl(s, KVM_GET_VCPU_MMAP_SIZE, 0);
248 DPRINTF("KVM_GET_VCPU_MMAP_SIZE failed\n");
252 cpu->kvm_run = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE, MAP_SHARED,
254 if (cpu->kvm_run == MAP_FAILED) {
256 DPRINTF("mmap'ing vcpu state failed\n");
260 if (s->coalesced_mmio && !s->coalesced_mmio_ring) {
261 s->coalesced_mmio_ring =
262 (void *)cpu->kvm_run + s->coalesced_mmio * PAGE_SIZE;
265 ret = kvm_arch_init_vcpu(cpu);
271 * dirty pages logging control
274 static int kvm_mem_flags(KVMState *s, bool log_dirty, bool readonly)
277 flags = log_dirty ? KVM_MEM_LOG_DIRTY_PAGES : 0;
278 if (readonly && kvm_readonly_mem_allowed) {
279 flags |= KVM_MEM_READONLY;
284 static int kvm_slot_dirty_pages_log_change(KVMSlot *mem, bool log_dirty)
286 KVMState *s = kvm_state;
287 int flags, mask = KVM_MEM_LOG_DIRTY_PAGES;
290 old_flags = mem->flags;
292 flags = (mem->flags & ~mask) | kvm_mem_flags(s, log_dirty, false);
295 /* If nothing changed effectively, no need to issue ioctl */
296 if (s->migration_log) {
297 flags |= KVM_MEM_LOG_DIRTY_PAGES;
300 if (flags == old_flags) {
304 return kvm_set_user_memory_region(s, mem);
307 static int kvm_dirty_pages_log_change(hwaddr phys_addr,
308 ram_addr_t size, bool log_dirty)
310 KVMState *s = kvm_state;
311 KVMSlot *mem = kvm_lookup_matching_slot(s, phys_addr, phys_addr + size);
314 fprintf(stderr, "BUG: %s: invalid parameters " TARGET_FMT_plx "-"
315 TARGET_FMT_plx "\n", __func__, phys_addr,
316 (hwaddr)(phys_addr + size - 1));
319 return kvm_slot_dirty_pages_log_change(mem, log_dirty);
322 static void kvm_log_start(MemoryListener *listener,
323 MemoryRegionSection *section)
327 r = kvm_dirty_pages_log_change(section->offset_within_address_space,
328 int128_get64(section->size), true);
334 static void kvm_log_stop(MemoryListener *listener,
335 MemoryRegionSection *section)
339 r = kvm_dirty_pages_log_change(section->offset_within_address_space,
340 int128_get64(section->size), false);
346 static int kvm_set_migration_log(int enable)
348 KVMState *s = kvm_state;
352 s->migration_log = enable;
354 for (i = 0; i < s->nr_slots; i++) {
357 if (!mem->memory_size) {
360 if (!!(mem->flags & KVM_MEM_LOG_DIRTY_PAGES) == enable) {
363 err = kvm_set_user_memory_region(s, mem);
371 /* get kvm's dirty pages bitmap and update qemu's */
372 static int kvm_get_dirty_pages_log_range(MemoryRegionSection *section,
373 unsigned long *bitmap)
375 ram_addr_t start = section->offset_within_region + section->mr->ram_addr;
376 ram_addr_t pages = int128_get64(section->size) / getpagesize();
378 cpu_physical_memory_set_dirty_lebitmap(bitmap, start, pages);
382 #define ALIGN(x, y) (((x)+(y)-1) & ~((y)-1))
385 * kvm_physical_sync_dirty_bitmap - Grab dirty bitmap from kernel space
386 * This function updates qemu's dirty bitmap using
387 * memory_region_set_dirty(). This means all bits are set
390 * @start_add: start of logged region.
391 * @end_addr: end of logged region.
393 static int kvm_physical_sync_dirty_bitmap(MemoryRegionSection *section)
395 KVMState *s = kvm_state;
396 unsigned long size, allocated_size = 0;
400 hwaddr start_addr = section->offset_within_address_space;
401 hwaddr end_addr = start_addr + int128_get64(section->size);
403 d.dirty_bitmap = NULL;
404 while (start_addr < end_addr) {
405 mem = kvm_lookup_overlapping_slot(s, start_addr, end_addr);
410 /* XXX bad kernel interface alert
411 * For dirty bitmap, kernel allocates array of size aligned to
412 * bits-per-long. But for case when the kernel is 64bits and
413 * the userspace is 32bits, userspace can't align to the same
414 * bits-per-long, since sizeof(long) is different between kernel
415 * and user space. This way, userspace will provide buffer which
416 * may be 4 bytes less than the kernel will use, resulting in
417 * userspace memory corruption (which is not detectable by valgrind
418 * too, in most cases).
419 * So for now, let's align to 64 instead of HOST_LONG_BITS here, in
420 * a hope that sizeof(long) wont become >8 any time soon.
422 size = ALIGN(((mem->memory_size) >> TARGET_PAGE_BITS),
423 /*HOST_LONG_BITS*/ 64) / 8;
424 if (!d.dirty_bitmap) {
425 d.dirty_bitmap = g_malloc(size);
426 } else if (size > allocated_size) {
427 d.dirty_bitmap = g_realloc(d.dirty_bitmap, size);
429 allocated_size = size;
430 memset(d.dirty_bitmap, 0, allocated_size);
434 if (kvm_vm_ioctl(s, KVM_GET_DIRTY_LOG, &d) == -1) {
435 DPRINTF("ioctl failed %d\n", errno);
440 kvm_get_dirty_pages_log_range(section, d.dirty_bitmap);
441 start_addr = mem->start_addr + mem->memory_size;
443 g_free(d.dirty_bitmap);
448 static void kvm_coalesce_mmio_region(MemoryListener *listener,
449 MemoryRegionSection *secion,
450 hwaddr start, hwaddr size)
452 KVMState *s = kvm_state;
454 if (s->coalesced_mmio) {
455 struct kvm_coalesced_mmio_zone zone;
461 (void)kvm_vm_ioctl(s, KVM_REGISTER_COALESCED_MMIO, &zone);
465 static void kvm_uncoalesce_mmio_region(MemoryListener *listener,
466 MemoryRegionSection *secion,
467 hwaddr start, hwaddr size)
469 KVMState *s = kvm_state;
471 if (s->coalesced_mmio) {
472 struct kvm_coalesced_mmio_zone zone;
478 (void)kvm_vm_ioctl(s, KVM_UNREGISTER_COALESCED_MMIO, &zone);
482 int kvm_check_extension(KVMState *s, unsigned int extension)
486 ret = kvm_ioctl(s, KVM_CHECK_EXTENSION, extension);
494 static int kvm_set_ioeventfd_mmio(int fd, hwaddr addr, uint32_t val,
495 bool assign, uint32_t size, bool datamatch)
498 struct kvm_ioeventfd iofd;
500 iofd.datamatch = datamatch ? val : 0;
506 if (!kvm_enabled()) {
511 iofd.flags |= KVM_IOEVENTFD_FLAG_DATAMATCH;
514 iofd.flags |= KVM_IOEVENTFD_FLAG_DEASSIGN;
517 ret = kvm_vm_ioctl(kvm_state, KVM_IOEVENTFD, &iofd);
526 static int kvm_set_ioeventfd_pio(int fd, uint16_t addr, uint16_t val,
527 bool assign, uint32_t size, bool datamatch)
529 struct kvm_ioeventfd kick = {
530 .datamatch = datamatch ? val : 0,
532 .flags = KVM_IOEVENTFD_FLAG_PIO,
537 if (!kvm_enabled()) {
541 kick.flags |= KVM_IOEVENTFD_FLAG_DATAMATCH;
544 kick.flags |= KVM_IOEVENTFD_FLAG_DEASSIGN;
546 r = kvm_vm_ioctl(kvm_state, KVM_IOEVENTFD, &kick);
554 static int kvm_check_many_ioeventfds(void)
556 /* Userspace can use ioeventfd for io notification. This requires a host
557 * that supports eventfd(2) and an I/O thread; since eventfd does not
558 * support SIGIO it cannot interrupt the vcpu.
560 * Older kernels have a 6 device limit on the KVM io bus. Find out so we
561 * can avoid creating too many ioeventfds.
563 #if defined(CONFIG_EVENTFD)
566 for (i = 0; i < ARRAY_SIZE(ioeventfds); i++) {
567 ioeventfds[i] = eventfd(0, EFD_CLOEXEC);
568 if (ioeventfds[i] < 0) {
571 ret = kvm_set_ioeventfd_pio(ioeventfds[i], 0, i, true, 2, true);
573 close(ioeventfds[i]);
578 /* Decide whether many devices are supported or not */
579 ret = i == ARRAY_SIZE(ioeventfds);
582 kvm_set_ioeventfd_pio(ioeventfds[i], 0, i, false, 2, true);
583 close(ioeventfds[i]);
591 static const KVMCapabilityInfo *
592 kvm_check_extension_list(KVMState *s, const KVMCapabilityInfo *list)
595 if (!kvm_check_extension(s, list->value)) {
603 static void kvm_set_phys_mem(MemoryRegionSection *section, bool add)
605 KVMState *s = kvm_state;
608 MemoryRegion *mr = section->mr;
609 bool log_dirty = memory_region_is_logging(mr);
610 bool writeable = !mr->readonly && !mr->rom_device;
611 bool readonly_flag = mr->readonly || memory_region_is_romd(mr);
612 hwaddr start_addr = section->offset_within_address_space;
613 ram_addr_t size = int128_get64(section->size);
617 /* kvm works in page size chunks, but the function may be called
618 with sub-page size and unaligned start address. */
619 delta = TARGET_PAGE_ALIGN(size) - size;
625 size &= TARGET_PAGE_MASK;
626 if (!size || (start_addr & ~TARGET_PAGE_MASK)) {
630 if (!memory_region_is_ram(mr)) {
631 if (writeable || !kvm_readonly_mem_allowed) {
633 } else if (!mr->romd_mode) {
634 /* If the memory device is not in romd_mode, then we actually want
635 * to remove the kvm memory slot so all accesses will trap. */
640 ram = memory_region_get_ram_ptr(mr) + section->offset_within_region + delta;
643 mem = kvm_lookup_overlapping_slot(s, start_addr, start_addr + size);
648 if (add && start_addr >= mem->start_addr &&
649 (start_addr + size <= mem->start_addr + mem->memory_size) &&
650 (ram - start_addr == mem->ram - mem->start_addr)) {
651 /* The new slot fits into the existing one and comes with
652 * identical parameters - update flags and done. */
653 kvm_slot_dirty_pages_log_change(mem, log_dirty);
659 if (mem->flags & KVM_MEM_LOG_DIRTY_PAGES) {
660 kvm_physical_sync_dirty_bitmap(section);
663 /* unregister the overlapping slot */
664 mem->memory_size = 0;
665 err = kvm_set_user_memory_region(s, mem);
667 fprintf(stderr, "%s: error unregistering overlapping slot: %s\n",
668 __func__, strerror(-err));
672 /* Workaround for older KVM versions: we can't join slots, even not by
673 * unregistering the previous ones and then registering the larger
674 * slot. We have to maintain the existing fragmentation. Sigh.
676 * This workaround assumes that the new slot starts at the same
677 * address as the first existing one. If not or if some overlapping
678 * slot comes around later, we will fail (not seen in practice so far)
679 * - and actually require a recent KVM version. */
680 if (s->broken_set_mem_region &&
681 old.start_addr == start_addr && old.memory_size < size && add) {
682 mem = kvm_alloc_slot(s);
683 mem->memory_size = old.memory_size;
684 mem->start_addr = old.start_addr;
686 mem->flags = kvm_mem_flags(s, log_dirty, readonly_flag);
688 err = kvm_set_user_memory_region(s, mem);
690 fprintf(stderr, "%s: error updating slot: %s\n", __func__,
695 start_addr += old.memory_size;
696 ram += old.memory_size;
697 size -= old.memory_size;
701 /* register prefix slot */
702 if (old.start_addr < start_addr) {
703 mem = kvm_alloc_slot(s);
704 mem->memory_size = start_addr - old.start_addr;
705 mem->start_addr = old.start_addr;
707 mem->flags = kvm_mem_flags(s, log_dirty, readonly_flag);
709 err = kvm_set_user_memory_region(s, mem);
711 fprintf(stderr, "%s: error registering prefix slot: %s\n",
712 __func__, strerror(-err));
714 fprintf(stderr, "%s: This is probably because your kernel's " \
715 "PAGE_SIZE is too big. Please try to use 4k " \
716 "PAGE_SIZE!\n", __func__);
722 /* register suffix slot */
723 if (old.start_addr + old.memory_size > start_addr + size) {
724 ram_addr_t size_delta;
726 mem = kvm_alloc_slot(s);
727 mem->start_addr = start_addr + size;
728 size_delta = mem->start_addr - old.start_addr;
729 mem->memory_size = old.memory_size - size_delta;
730 mem->ram = old.ram + size_delta;
731 mem->flags = kvm_mem_flags(s, log_dirty, readonly_flag);
733 err = kvm_set_user_memory_region(s, mem);
735 fprintf(stderr, "%s: error registering suffix slot: %s\n",
736 __func__, strerror(-err));
742 /* in case the KVM bug workaround already "consumed" the new slot */
749 mem = kvm_alloc_slot(s);
750 mem->memory_size = size;
751 mem->start_addr = start_addr;
753 mem->flags = kvm_mem_flags(s, log_dirty, readonly_flag);
755 err = kvm_set_user_memory_region(s, mem);
757 fprintf(stderr, "%s: error registering slot: %s\n", __func__,
763 static void kvm_region_add(MemoryListener *listener,
764 MemoryRegionSection *section)
766 memory_region_ref(section->mr);
767 kvm_set_phys_mem(section, true);
770 static void kvm_region_del(MemoryListener *listener,
771 MemoryRegionSection *section)
773 kvm_set_phys_mem(section, false);
774 memory_region_unref(section->mr);
777 static void kvm_log_sync(MemoryListener *listener,
778 MemoryRegionSection *section)
782 r = kvm_physical_sync_dirty_bitmap(section);
788 static void kvm_log_global_start(struct MemoryListener *listener)
792 r = kvm_set_migration_log(1);
796 static void kvm_log_global_stop(struct MemoryListener *listener)
800 r = kvm_set_migration_log(0);
804 static void kvm_mem_ioeventfd_add(MemoryListener *listener,
805 MemoryRegionSection *section,
806 bool match_data, uint64_t data,
809 int fd = event_notifier_get_fd(e);
812 r = kvm_set_ioeventfd_mmio(fd, section->offset_within_address_space,
813 data, true, int128_get64(section->size),
816 fprintf(stderr, "%s: error adding ioeventfd: %s\n",
817 __func__, strerror(-r));
822 static void kvm_mem_ioeventfd_del(MemoryListener *listener,
823 MemoryRegionSection *section,
824 bool match_data, uint64_t data,
827 int fd = event_notifier_get_fd(e);
830 r = kvm_set_ioeventfd_mmio(fd, section->offset_within_address_space,
831 data, false, int128_get64(section->size),
838 static void kvm_io_ioeventfd_add(MemoryListener *listener,
839 MemoryRegionSection *section,
840 bool match_data, uint64_t data,
843 int fd = event_notifier_get_fd(e);
846 r = kvm_set_ioeventfd_pio(fd, section->offset_within_address_space,
847 data, true, int128_get64(section->size),
850 fprintf(stderr, "%s: error adding ioeventfd: %s\n",
851 __func__, strerror(-r));
856 static void kvm_io_ioeventfd_del(MemoryListener *listener,
857 MemoryRegionSection *section,
858 bool match_data, uint64_t data,
862 int fd = event_notifier_get_fd(e);
865 r = kvm_set_ioeventfd_pio(fd, section->offset_within_address_space,
866 data, false, int128_get64(section->size),
873 static MemoryListener kvm_memory_listener = {
874 .region_add = kvm_region_add,
875 .region_del = kvm_region_del,
876 .log_start = kvm_log_start,
877 .log_stop = kvm_log_stop,
878 .log_sync = kvm_log_sync,
879 .log_global_start = kvm_log_global_start,
880 .log_global_stop = kvm_log_global_stop,
881 .eventfd_add = kvm_mem_ioeventfd_add,
882 .eventfd_del = kvm_mem_ioeventfd_del,
883 .coalesced_mmio_add = kvm_coalesce_mmio_region,
884 .coalesced_mmio_del = kvm_uncoalesce_mmio_region,
888 static MemoryListener kvm_io_listener = {
889 .eventfd_add = kvm_io_ioeventfd_add,
890 .eventfd_del = kvm_io_ioeventfd_del,
894 static void kvm_handle_interrupt(CPUState *cpu, int mask)
896 cpu->interrupt_request |= mask;
898 if (!qemu_cpu_is_self(cpu)) {
903 int kvm_set_irq(KVMState *s, int irq, int level)
905 struct kvm_irq_level event;
908 assert(kvm_async_interrupts_enabled());
912 ret = kvm_vm_ioctl(s, s->irq_set_ioctl, &event);
914 perror("kvm_set_irq");
918 return (s->irq_set_ioctl == KVM_IRQ_LINE) ? 1 : event.status;
921 #ifdef KVM_CAP_IRQ_ROUTING
922 typedef struct KVMMSIRoute {
923 struct kvm_irq_routing_entry kroute;
924 QTAILQ_ENTRY(KVMMSIRoute) entry;
927 static void set_gsi(KVMState *s, unsigned int gsi)
929 s->used_gsi_bitmap[gsi / 32] |= 1U << (gsi % 32);
932 static void clear_gsi(KVMState *s, unsigned int gsi)
934 s->used_gsi_bitmap[gsi / 32] &= ~(1U << (gsi % 32));
937 void kvm_init_irq_routing(KVMState *s)
941 gsi_count = kvm_check_extension(s, KVM_CAP_IRQ_ROUTING);
943 unsigned int gsi_bits, i;
945 /* Round up so we can search ints using ffs */
946 gsi_bits = ALIGN(gsi_count, 32);
947 s->used_gsi_bitmap = g_malloc0(gsi_bits / 8);
948 s->gsi_count = gsi_count;
950 /* Mark any over-allocated bits as already in use */
951 for (i = gsi_count; i < gsi_bits; i++) {
956 s->irq_routes = g_malloc0(sizeof(*s->irq_routes));
957 s->nr_allocated_irq_routes = 0;
959 if (!s->direct_msi) {
960 for (i = 0; i < KVM_MSI_HASHTAB_SIZE; i++) {
961 QTAILQ_INIT(&s->msi_hashtab[i]);
965 kvm_arch_init_irq_routing(s);
968 void kvm_irqchip_commit_routes(KVMState *s)
972 s->irq_routes->flags = 0;
973 ret = kvm_vm_ioctl(s, KVM_SET_GSI_ROUTING, s->irq_routes);
977 static void kvm_add_routing_entry(KVMState *s,
978 struct kvm_irq_routing_entry *entry)
980 struct kvm_irq_routing_entry *new;
983 if (s->irq_routes->nr == s->nr_allocated_irq_routes) {
984 n = s->nr_allocated_irq_routes * 2;
988 size = sizeof(struct kvm_irq_routing);
989 size += n * sizeof(*new);
990 s->irq_routes = g_realloc(s->irq_routes, size);
991 s->nr_allocated_irq_routes = n;
993 n = s->irq_routes->nr++;
994 new = &s->irq_routes->entries[n];
998 set_gsi(s, entry->gsi);
1001 static int kvm_update_routing_entry(KVMState *s,
1002 struct kvm_irq_routing_entry *new_entry)
1004 struct kvm_irq_routing_entry *entry;
1007 for (n = 0; n < s->irq_routes->nr; n++) {
1008 entry = &s->irq_routes->entries[n];
1009 if (entry->gsi != new_entry->gsi) {
1013 if(!memcmp(entry, new_entry, sizeof *entry)) {
1017 *entry = *new_entry;
1019 kvm_irqchip_commit_routes(s);
1027 void kvm_irqchip_add_irq_route(KVMState *s, int irq, int irqchip, int pin)
1029 struct kvm_irq_routing_entry e = {};
1031 assert(pin < s->gsi_count);
1034 e.type = KVM_IRQ_ROUTING_IRQCHIP;
1036 e.u.irqchip.irqchip = irqchip;
1037 e.u.irqchip.pin = pin;
1038 kvm_add_routing_entry(s, &e);
1041 void kvm_irqchip_release_virq(KVMState *s, int virq)
1043 struct kvm_irq_routing_entry *e;
1046 if (kvm_gsi_direct_mapping()) {
1050 for (i = 0; i < s->irq_routes->nr; i++) {
1051 e = &s->irq_routes->entries[i];
1052 if (e->gsi == virq) {
1053 s->irq_routes->nr--;
1054 *e = s->irq_routes->entries[s->irq_routes->nr];
1060 static unsigned int kvm_hash_msi(uint32_t data)
1062 /* This is optimized for IA32 MSI layout. However, no other arch shall
1063 * repeat the mistake of not providing a direct MSI injection API. */
1067 static void kvm_flush_dynamic_msi_routes(KVMState *s)
1069 KVMMSIRoute *route, *next;
1072 for (hash = 0; hash < KVM_MSI_HASHTAB_SIZE; hash++) {
1073 QTAILQ_FOREACH_SAFE(route, &s->msi_hashtab[hash], entry, next) {
1074 kvm_irqchip_release_virq(s, route->kroute.gsi);
1075 QTAILQ_REMOVE(&s->msi_hashtab[hash], route, entry);
1081 static int kvm_irqchip_get_virq(KVMState *s)
1083 uint32_t *word = s->used_gsi_bitmap;
1084 int max_words = ALIGN(s->gsi_count, 32) / 32;
1089 /* Return the lowest unused GSI in the bitmap */
1090 for (i = 0; i < max_words; i++) {
1091 bit = ffs(~word[i]);
1096 return bit - 1 + i * 32;
1098 if (!s->direct_msi && retry) {
1100 kvm_flush_dynamic_msi_routes(s);
1107 static KVMMSIRoute *kvm_lookup_msi_route(KVMState *s, MSIMessage msg)
1109 unsigned int hash = kvm_hash_msi(msg.data);
1112 QTAILQ_FOREACH(route, &s->msi_hashtab[hash], entry) {
1113 if (route->kroute.u.msi.address_lo == (uint32_t)msg.address &&
1114 route->kroute.u.msi.address_hi == (msg.address >> 32) &&
1115 route->kroute.u.msi.data == le32_to_cpu(msg.data)) {
1122 int kvm_irqchip_send_msi(KVMState *s, MSIMessage msg)
1127 if (s->direct_msi) {
1128 msi.address_lo = (uint32_t)msg.address;
1129 msi.address_hi = msg.address >> 32;
1130 msi.data = le32_to_cpu(msg.data);
1132 memset(msi.pad, 0, sizeof(msi.pad));
1134 return kvm_vm_ioctl(s, KVM_SIGNAL_MSI, &msi);
1137 route = kvm_lookup_msi_route(s, msg);
1141 virq = kvm_irqchip_get_virq(s);
1146 route = g_malloc0(sizeof(KVMMSIRoute));
1147 route->kroute.gsi = virq;
1148 route->kroute.type = KVM_IRQ_ROUTING_MSI;
1149 route->kroute.flags = 0;
1150 route->kroute.u.msi.address_lo = (uint32_t)msg.address;
1151 route->kroute.u.msi.address_hi = msg.address >> 32;
1152 route->kroute.u.msi.data = le32_to_cpu(msg.data);
1154 kvm_add_routing_entry(s, &route->kroute);
1155 kvm_irqchip_commit_routes(s);
1157 QTAILQ_INSERT_TAIL(&s->msi_hashtab[kvm_hash_msi(msg.data)], route,
1161 assert(route->kroute.type == KVM_IRQ_ROUTING_MSI);
1163 return kvm_set_irq(s, route->kroute.gsi, 1);
1166 int kvm_irqchip_add_msi_route(KVMState *s, MSIMessage msg)
1168 struct kvm_irq_routing_entry kroute = {};
1171 if (kvm_gsi_direct_mapping()) {
1172 return msg.data & 0xffff;
1175 if (!kvm_gsi_routing_enabled()) {
1179 virq = kvm_irqchip_get_virq(s);
1185 kroute.type = KVM_IRQ_ROUTING_MSI;
1187 kroute.u.msi.address_lo = (uint32_t)msg.address;
1188 kroute.u.msi.address_hi = msg.address >> 32;
1189 kroute.u.msi.data = le32_to_cpu(msg.data);
1191 kvm_add_routing_entry(s, &kroute);
1192 kvm_irqchip_commit_routes(s);
1197 int kvm_irqchip_update_msi_route(KVMState *s, int virq, MSIMessage msg)
1199 struct kvm_irq_routing_entry kroute = {};
1201 if (kvm_gsi_direct_mapping()) {
1205 if (!kvm_irqchip_in_kernel()) {
1210 kroute.type = KVM_IRQ_ROUTING_MSI;
1212 kroute.u.msi.address_lo = (uint32_t)msg.address;
1213 kroute.u.msi.address_hi = msg.address >> 32;
1214 kroute.u.msi.data = le32_to_cpu(msg.data);
1216 return kvm_update_routing_entry(s, &kroute);
1219 static int kvm_irqchip_assign_irqfd(KVMState *s, int fd, int rfd, int virq,
1222 struct kvm_irqfd irqfd = {
1225 .flags = assign ? 0 : KVM_IRQFD_FLAG_DEASSIGN,
1229 irqfd.flags |= KVM_IRQFD_FLAG_RESAMPLE;
1230 irqfd.resamplefd = rfd;
1233 if (!kvm_irqfds_enabled()) {
1237 return kvm_vm_ioctl(s, KVM_IRQFD, &irqfd);
1240 int kvm_irqchip_add_adapter_route(KVMState *s, AdapterInfo *adapter)
1242 struct kvm_irq_routing_entry kroute;
1245 if (!kvm_gsi_routing_enabled()) {
1249 virq = kvm_irqchip_get_virq(s);
1255 kroute.type = KVM_IRQ_ROUTING_S390_ADAPTER;
1257 kroute.u.adapter.summary_addr = adapter->summary_addr;
1258 kroute.u.adapter.ind_addr = adapter->ind_addr;
1259 kroute.u.adapter.summary_offset = adapter->summary_offset;
1260 kroute.u.adapter.ind_offset = adapter->ind_offset;
1261 kroute.u.adapter.adapter_id = adapter->adapter_id;
1263 kvm_add_routing_entry(s, &kroute);
1264 kvm_irqchip_commit_routes(s);
1269 #else /* !KVM_CAP_IRQ_ROUTING */
1271 void kvm_init_irq_routing(KVMState *s)
1275 void kvm_irqchip_release_virq(KVMState *s, int virq)
1279 int kvm_irqchip_send_msi(KVMState *s, MSIMessage msg)
1284 int kvm_irqchip_add_msi_route(KVMState *s, MSIMessage msg)
1289 int kvm_irqchip_add_adapter_route(KVMState *s, AdapterInfo *adapter)
1294 static int kvm_irqchip_assign_irqfd(KVMState *s, int fd, int virq, bool assign)
1299 int kvm_irqchip_update_msi_route(KVMState *s, int virq, MSIMessage msg)
1303 #endif /* !KVM_CAP_IRQ_ROUTING */
1305 int kvm_irqchip_add_irqfd_notifier(KVMState *s, EventNotifier *n,
1306 EventNotifier *rn, int virq)
1308 return kvm_irqchip_assign_irqfd(s, event_notifier_get_fd(n),
1309 rn ? event_notifier_get_fd(rn) : -1, virq, true);
1312 int kvm_irqchip_remove_irqfd_notifier(KVMState *s, EventNotifier *n, int virq)
1314 return kvm_irqchip_assign_irqfd(s, event_notifier_get_fd(n), -1, virq,
1318 static int kvm_irqchip_create(KVMState *s)
1322 if (!qemu_opt_get_bool(qemu_get_machine_opts(), "kernel_irqchip", true) ||
1323 (!kvm_check_extension(s, KVM_CAP_IRQCHIP) &&
1324 (kvm_vm_enable_cap(s, KVM_CAP_S390_IRQCHIP, 0) < 0))) {
1328 /* First probe and see if there's a arch-specific hook to create the
1329 * in-kernel irqchip for us */
1330 ret = kvm_arch_irqchip_create(s);
1333 } else if (ret == 0) {
1334 ret = kvm_vm_ioctl(s, KVM_CREATE_IRQCHIP);
1336 fprintf(stderr, "Create kernel irqchip failed\n");
1341 kvm_kernel_irqchip = true;
1342 /* If we have an in-kernel IRQ chip then we must have asynchronous
1343 * interrupt delivery (though the reverse is not necessarily true)
1345 kvm_async_interrupts_allowed = true;
1346 kvm_halt_in_kernel_allowed = true;
1348 kvm_init_irq_routing(s);
1353 /* Find number of supported CPUs using the recommended
1354 * procedure from the kernel API documentation to cope with
1355 * older kernels that may be missing capabilities.
1357 static int kvm_recommended_vcpus(KVMState *s)
1359 int ret = kvm_check_extension(s, KVM_CAP_NR_VCPUS);
1360 return (ret) ? ret : 4;
1363 static int kvm_max_vcpus(KVMState *s)
1365 int ret = kvm_check_extension(s, KVM_CAP_MAX_VCPUS);
1366 return (ret) ? ret : kvm_recommended_vcpus(s);
1369 int kvm_init(MachineClass *mc)
1371 static const char upgrade_note[] =
1372 "Please upgrade to at least kernel 2.6.29 or recent kvm-kmod\n"
1373 "(see http://sourceforge.net/projects/kvm).\n";
1378 { "SMP", smp_cpus },
1379 { "hotpluggable", max_cpus },
1382 int soft_vcpus_limit, hard_vcpus_limit;
1384 const KVMCapabilityInfo *missing_cap;
1387 const char *kvm_type;
1389 s = g_malloc0(sizeof(KVMState));
1392 * On systems where the kernel can support different base page
1393 * sizes, host page size may be different from TARGET_PAGE_SIZE,
1394 * even with KVM. TARGET_PAGE_SIZE is assumed to be the minimum
1395 * page size for the system though.
1397 assert(TARGET_PAGE_SIZE <= getpagesize());
1400 #ifdef KVM_CAP_SET_GUEST_DEBUG
1401 QTAILQ_INIT(&s->kvm_sw_breakpoints);
1404 s->fd = qemu_open("/dev/kvm", O_RDWR);
1406 fprintf(stderr, "Could not access KVM kernel module: %m\n");
1411 ret = kvm_ioctl(s, KVM_GET_API_VERSION, 0);
1412 if (ret < KVM_API_VERSION) {
1416 fprintf(stderr, "kvm version too old\n");
1420 if (ret > KVM_API_VERSION) {
1422 fprintf(stderr, "kvm version not supported\n");
1426 s->nr_slots = kvm_check_extension(s, KVM_CAP_NR_MEMSLOTS);
1428 /* If unspecified, use the default value */
1433 s->slots = g_malloc0(s->nr_slots * sizeof(KVMSlot));
1435 for (i = 0; i < s->nr_slots; i++) {
1436 s->slots[i].slot = i;
1439 /* check the vcpu limits */
1440 soft_vcpus_limit = kvm_recommended_vcpus(s);
1441 hard_vcpus_limit = kvm_max_vcpus(s);
1444 if (nc->num > soft_vcpus_limit) {
1446 "Warning: Number of %s cpus requested (%d) exceeds "
1447 "the recommended cpus supported by KVM (%d)\n",
1448 nc->name, nc->num, soft_vcpus_limit);
1450 if (nc->num > hard_vcpus_limit) {
1451 fprintf(stderr, "Number of %s cpus requested (%d) exceeds "
1452 "the maximum cpus supported by KVM (%d)\n",
1453 nc->name, nc->num, hard_vcpus_limit);
1460 kvm_type = qemu_opt_get(qemu_get_machine_opts(), "kvm-type");
1462 type = mc->kvm_type(kvm_type);
1463 } else if (kvm_type) {
1465 fprintf(stderr, "Invalid argument kvm-type=%s\n", kvm_type);
1470 ret = kvm_ioctl(s, KVM_CREATE_VM, type);
1471 } while (ret == -EINTR);
1474 fprintf(stderr, "ioctl(KVM_CREATE_VM) failed: %d %s\n", -ret,
1478 fprintf(stderr, "Please add the 'switch_amode' kernel parameter to "
1479 "your host kernel command line\n");
1485 missing_cap = kvm_check_extension_list(s, kvm_required_capabilites);
1488 kvm_check_extension_list(s, kvm_arch_required_capabilities);
1492 fprintf(stderr, "kvm does not support %s\n%s",
1493 missing_cap->name, upgrade_note);
1497 s->coalesced_mmio = kvm_check_extension(s, KVM_CAP_COALESCED_MMIO);
1499 s->broken_set_mem_region = 1;
1500 ret = kvm_check_extension(s, KVM_CAP_JOIN_MEMORY_REGIONS_WORKS);
1502 s->broken_set_mem_region = 0;
1505 #ifdef KVM_CAP_VCPU_EVENTS
1506 s->vcpu_events = kvm_check_extension(s, KVM_CAP_VCPU_EVENTS);
1509 s->robust_singlestep =
1510 kvm_check_extension(s, KVM_CAP_X86_ROBUST_SINGLESTEP);
1512 #ifdef KVM_CAP_DEBUGREGS
1513 s->debugregs = kvm_check_extension(s, KVM_CAP_DEBUGREGS);
1516 #ifdef KVM_CAP_XSAVE
1517 s->xsave = kvm_check_extension(s, KVM_CAP_XSAVE);
1521 s->xcrs = kvm_check_extension(s, KVM_CAP_XCRS);
1524 #ifdef KVM_CAP_PIT_STATE2
1525 s->pit_state2 = kvm_check_extension(s, KVM_CAP_PIT_STATE2);
1528 #ifdef KVM_CAP_IRQ_ROUTING
1529 s->direct_msi = (kvm_check_extension(s, KVM_CAP_SIGNAL_MSI) > 0);
1532 s->intx_set_mask = kvm_check_extension(s, KVM_CAP_PCI_2_3);
1534 s->irq_set_ioctl = KVM_IRQ_LINE;
1535 if (kvm_check_extension(s, KVM_CAP_IRQ_INJECT_STATUS)) {
1536 s->irq_set_ioctl = KVM_IRQ_LINE_STATUS;
1539 #ifdef KVM_CAP_READONLY_MEM
1540 kvm_readonly_mem_allowed =
1541 (kvm_check_extension(s, KVM_CAP_READONLY_MEM) > 0);
1544 ret = kvm_arch_init(s);
1549 ret = kvm_irqchip_create(s);
1555 memory_listener_register(&kvm_memory_listener, &address_space_memory);
1556 memory_listener_register(&kvm_io_listener, &address_space_io);
1558 s->many_ioeventfds = kvm_check_many_ioeventfds();
1560 cpu_interrupt_handler = kvm_handle_interrupt;
1578 static void kvm_handle_io(uint16_t port, void *data, int direction, int size,
1582 uint8_t *ptr = data;
1584 for (i = 0; i < count; i++) {
1585 address_space_rw(&address_space_io, port, ptr, size,
1586 direction == KVM_EXIT_IO_OUT);
1591 static int kvm_handle_internal_error(CPUState *cpu, struct kvm_run *run)
1593 fprintf(stderr, "KVM internal error. Suberror: %d\n",
1594 run->internal.suberror);
1596 if (kvm_check_extension(kvm_state, KVM_CAP_INTERNAL_ERROR_DATA)) {
1599 for (i = 0; i < run->internal.ndata; ++i) {
1600 fprintf(stderr, "extra data[%d]: %"PRIx64"\n",
1601 i, (uint64_t)run->internal.data[i]);
1604 if (run->internal.suberror == KVM_INTERNAL_ERROR_EMULATION) {
1605 fprintf(stderr, "emulation failure\n");
1606 if (!kvm_arch_stop_on_emulation_error(cpu)) {
1607 cpu_dump_state(cpu, stderr, fprintf, CPU_DUMP_CODE);
1608 return EXCP_INTERRUPT;
1611 /* FIXME: Should trigger a qmp message to let management know
1612 * something went wrong.
1617 void kvm_flush_coalesced_mmio_buffer(void)
1619 KVMState *s = kvm_state;
1621 if (s->coalesced_flush_in_progress) {
1625 s->coalesced_flush_in_progress = true;
1627 if (s->coalesced_mmio_ring) {
1628 struct kvm_coalesced_mmio_ring *ring = s->coalesced_mmio_ring;
1629 while (ring->first != ring->last) {
1630 struct kvm_coalesced_mmio *ent;
1632 ent = &ring->coalesced_mmio[ring->first];
1634 cpu_physical_memory_write(ent->phys_addr, ent->data, ent->len);
1636 ring->first = (ring->first + 1) % KVM_COALESCED_MMIO_MAX;
1640 s->coalesced_flush_in_progress = false;
1643 static void do_kvm_cpu_synchronize_state(void *arg)
1645 CPUState *cpu = arg;
1647 if (!cpu->kvm_vcpu_dirty) {
1648 kvm_arch_get_registers(cpu);
1649 cpu->kvm_vcpu_dirty = true;
1653 void kvm_cpu_synchronize_state(CPUState *cpu)
1655 if (!cpu->kvm_vcpu_dirty) {
1656 run_on_cpu(cpu, do_kvm_cpu_synchronize_state, cpu);
1660 void kvm_cpu_synchronize_post_reset(CPUState *cpu)
1662 kvm_arch_put_registers(cpu, KVM_PUT_RESET_STATE);
1663 cpu->kvm_vcpu_dirty = false;
1666 void kvm_cpu_synchronize_post_init(CPUState *cpu)
1668 kvm_arch_put_registers(cpu, KVM_PUT_FULL_STATE);
1669 cpu->kvm_vcpu_dirty = false;
1672 int kvm_cpu_exec(CPUState *cpu)
1674 struct kvm_run *run = cpu->kvm_run;
1677 DPRINTF("kvm_cpu_exec()\n");
1679 if (kvm_arch_process_async_events(cpu)) {
1680 cpu->exit_request = 0;
1685 if (cpu->kvm_vcpu_dirty) {
1686 kvm_arch_put_registers(cpu, KVM_PUT_RUNTIME_STATE);
1687 cpu->kvm_vcpu_dirty = false;
1690 kvm_arch_pre_run(cpu, run);
1691 if (cpu->exit_request) {
1692 DPRINTF("interrupt exit requested\n");
1694 * KVM requires us to reenter the kernel after IO exits to complete
1695 * instruction emulation. This self-signal will ensure that we
1698 qemu_cpu_kick_self();
1700 qemu_mutex_unlock_iothread();
1702 run_ret = kvm_vcpu_ioctl(cpu, KVM_RUN, 0);
1704 qemu_mutex_lock_iothread();
1705 kvm_arch_post_run(cpu, run);
1708 if (run_ret == -EINTR || run_ret == -EAGAIN) {
1709 DPRINTF("io window exit\n");
1710 ret = EXCP_INTERRUPT;
1713 fprintf(stderr, "error: kvm run failed %s\n",
1714 strerror(-run_ret));
1718 trace_kvm_run_exit(cpu->cpu_index, run->exit_reason);
1719 switch (run->exit_reason) {
1721 DPRINTF("handle_io\n");
1722 kvm_handle_io(run->io.port,
1723 (uint8_t *)run + run->io.data_offset,
1730 DPRINTF("handle_mmio\n");
1731 cpu_physical_memory_rw(run->mmio.phys_addr,
1734 run->mmio.is_write);
1737 case KVM_EXIT_IRQ_WINDOW_OPEN:
1738 DPRINTF("irq_window_open\n");
1739 ret = EXCP_INTERRUPT;
1741 case KVM_EXIT_SHUTDOWN:
1742 DPRINTF("shutdown\n");
1743 qemu_system_reset_request();
1744 ret = EXCP_INTERRUPT;
1746 case KVM_EXIT_UNKNOWN:
1747 fprintf(stderr, "KVM: unknown exit, hardware reason %" PRIx64 "\n",
1748 (uint64_t)run->hw.hardware_exit_reason);
1751 case KVM_EXIT_INTERNAL_ERROR:
1752 ret = kvm_handle_internal_error(cpu, run);
1755 DPRINTF("kvm_arch_handle_exit\n");
1756 ret = kvm_arch_handle_exit(cpu, run);
1762 cpu_dump_state(cpu, stderr, fprintf, CPU_DUMP_CODE);
1763 vm_stop(RUN_STATE_INTERNAL_ERROR);
1766 cpu->exit_request = 0;
1770 int kvm_ioctl(KVMState *s, int type, ...)
1777 arg = va_arg(ap, void *);
1780 trace_kvm_ioctl(type, arg);
1781 ret = ioctl(s->fd, type, arg);
1788 int kvm_vm_ioctl(KVMState *s, int type, ...)
1795 arg = va_arg(ap, void *);
1798 trace_kvm_vm_ioctl(type, arg);
1799 ret = ioctl(s->vmfd, type, arg);
1806 int kvm_vcpu_ioctl(CPUState *cpu, int type, ...)
1813 arg = va_arg(ap, void *);
1816 trace_kvm_vcpu_ioctl(cpu->cpu_index, type, arg);
1817 ret = ioctl(cpu->kvm_fd, type, arg);
1824 int kvm_device_ioctl(int fd, int type, ...)
1831 arg = va_arg(ap, void *);
1834 trace_kvm_device_ioctl(fd, type, arg);
1835 ret = ioctl(fd, type, arg);
1842 int kvm_has_sync_mmu(void)
1844 return kvm_check_extension(kvm_state, KVM_CAP_SYNC_MMU);
1847 int kvm_has_vcpu_events(void)
1849 return kvm_state->vcpu_events;
1852 int kvm_has_robust_singlestep(void)
1854 return kvm_state->robust_singlestep;
1857 int kvm_has_debugregs(void)
1859 return kvm_state->debugregs;
1862 int kvm_has_xsave(void)
1864 return kvm_state->xsave;
1867 int kvm_has_xcrs(void)
1869 return kvm_state->xcrs;
1872 int kvm_has_pit_state2(void)
1874 return kvm_state->pit_state2;
1877 int kvm_has_many_ioeventfds(void)
1879 if (!kvm_enabled()) {
1882 return kvm_state->many_ioeventfds;
1885 int kvm_has_gsi_routing(void)
1887 #ifdef KVM_CAP_IRQ_ROUTING
1888 return kvm_check_extension(kvm_state, KVM_CAP_IRQ_ROUTING);
1894 int kvm_has_intx_set_mask(void)
1896 return kvm_state->intx_set_mask;
1899 void kvm_setup_guest_memory(void *start, size_t size)
1901 #ifdef CONFIG_VALGRIND_H
1902 VALGRIND_MAKE_MEM_DEFINED(start, size);
1904 if (!kvm_has_sync_mmu()) {
1905 int ret = qemu_madvise(start, size, QEMU_MADV_DONTFORK);
1908 perror("qemu_madvise");
1910 "Need MADV_DONTFORK in absence of synchronous KVM MMU\n");
1916 #ifdef KVM_CAP_SET_GUEST_DEBUG
1917 struct kvm_sw_breakpoint *kvm_find_sw_breakpoint(CPUState *cpu,
1920 struct kvm_sw_breakpoint *bp;
1922 QTAILQ_FOREACH(bp, &cpu->kvm_state->kvm_sw_breakpoints, entry) {
1930 int kvm_sw_breakpoints_active(CPUState *cpu)
1932 return !QTAILQ_EMPTY(&cpu->kvm_state->kvm_sw_breakpoints);
1935 struct kvm_set_guest_debug_data {
1936 struct kvm_guest_debug dbg;
1941 static void kvm_invoke_set_guest_debug(void *data)
1943 struct kvm_set_guest_debug_data *dbg_data = data;
1945 dbg_data->err = kvm_vcpu_ioctl(dbg_data->cpu, KVM_SET_GUEST_DEBUG,
1949 int kvm_update_guest_debug(CPUState *cpu, unsigned long reinject_trap)
1951 struct kvm_set_guest_debug_data data;
1953 data.dbg.control = reinject_trap;
1955 if (cpu->singlestep_enabled) {
1956 data.dbg.control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_SINGLESTEP;
1958 kvm_arch_update_guest_debug(cpu, &data.dbg);
1961 run_on_cpu(cpu, kvm_invoke_set_guest_debug, &data);
1965 int kvm_insert_breakpoint(CPUState *cpu, target_ulong addr,
1966 target_ulong len, int type)
1968 struct kvm_sw_breakpoint *bp;
1971 if (type == GDB_BREAKPOINT_SW) {
1972 bp = kvm_find_sw_breakpoint(cpu, addr);
1978 bp = g_malloc(sizeof(struct kvm_sw_breakpoint));
1985 err = kvm_arch_insert_sw_breakpoint(cpu, bp);
1991 QTAILQ_INSERT_HEAD(&cpu->kvm_state->kvm_sw_breakpoints, bp, entry);
1993 err = kvm_arch_insert_hw_breakpoint(addr, len, type);
2000 err = kvm_update_guest_debug(cpu, 0);
2008 int kvm_remove_breakpoint(CPUState *cpu, target_ulong addr,
2009 target_ulong len, int type)
2011 struct kvm_sw_breakpoint *bp;
2014 if (type == GDB_BREAKPOINT_SW) {
2015 bp = kvm_find_sw_breakpoint(cpu, addr);
2020 if (bp->use_count > 1) {
2025 err = kvm_arch_remove_sw_breakpoint(cpu, bp);
2030 QTAILQ_REMOVE(&cpu->kvm_state->kvm_sw_breakpoints, bp, entry);
2033 err = kvm_arch_remove_hw_breakpoint(addr, len, type);
2040 err = kvm_update_guest_debug(cpu, 0);
2048 void kvm_remove_all_breakpoints(CPUState *cpu)
2050 struct kvm_sw_breakpoint *bp, *next;
2051 KVMState *s = cpu->kvm_state;
2053 QTAILQ_FOREACH_SAFE(bp, &s->kvm_sw_breakpoints, entry, next) {
2054 if (kvm_arch_remove_sw_breakpoint(cpu, bp) != 0) {
2055 /* Try harder to find a CPU that currently sees the breakpoint. */
2057 if (kvm_arch_remove_sw_breakpoint(cpu, bp) == 0) {
2062 QTAILQ_REMOVE(&s->kvm_sw_breakpoints, bp, entry);
2065 kvm_arch_remove_all_hw_breakpoints();
2068 kvm_update_guest_debug(cpu, 0);
2072 #else /* !KVM_CAP_SET_GUEST_DEBUG */
2074 int kvm_update_guest_debug(CPUState *cpu, unsigned long reinject_trap)
2079 int kvm_insert_breakpoint(CPUState *cpu, target_ulong addr,
2080 target_ulong len, int type)
2085 int kvm_remove_breakpoint(CPUState *cpu, target_ulong addr,
2086 target_ulong len, int type)
2091 void kvm_remove_all_breakpoints(CPUState *cpu)
2094 #endif /* !KVM_CAP_SET_GUEST_DEBUG */
2096 int kvm_set_signal_mask(CPUState *cpu, const sigset_t *sigset)
2098 struct kvm_signal_mask *sigmask;
2102 return kvm_vcpu_ioctl(cpu, KVM_SET_SIGNAL_MASK, NULL);
2105 sigmask = g_malloc(sizeof(*sigmask) + sizeof(*sigset));
2108 memcpy(sigmask->sigset, sigset, sizeof(*sigset));
2109 r = kvm_vcpu_ioctl(cpu, KVM_SET_SIGNAL_MASK, sigmask);
2114 int kvm_on_sigbus_vcpu(CPUState *cpu, int code, void *addr)
2116 return kvm_arch_on_sigbus_vcpu(cpu, code, addr);
2119 int kvm_on_sigbus(int code, void *addr)
2121 return kvm_arch_on_sigbus(code, addr);
2124 int kvm_create_device(KVMState *s, uint64_t type, bool test)
2127 struct kvm_create_device create_dev;
2129 create_dev.type = type;
2131 create_dev.flags = test ? KVM_CREATE_DEVICE_TEST : 0;
2133 if (!kvm_check_extension(s, KVM_CAP_DEVICE_CTRL)) {
2137 ret = kvm_vm_ioctl(s, KVM_CREATE_DEVICE, &create_dev);
2142 return test ? 0 : create_dev.fd;
2145 int kvm_set_one_reg(CPUState *cs, uint64_t id, void *source)
2147 struct kvm_one_reg reg;
2151 reg.addr = (uintptr_t) source;
2152 r = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, ®);
2154 trace_kvm_failed_reg_set(id, strerror(r));
2159 int kvm_get_one_reg(CPUState *cs, uint64_t id, void *target)
2161 struct kvm_one_reg reg;
2165 reg.addr = (uintptr_t) target;
2166 r = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, ®);
2168 trace_kvm_failed_reg_get(id, strerror(r));